I'll add to @Tom Leek's answer with a bit more specifics on how one of the most common open spamfilter on the market (spamAssassin) checks whether something is spam. At the core of it, it is a simple set of rules (the default ones for the latest spamd version can be found at http://spamassassin.apache.org/tests_3_3_x.html ), some with very simplistic checks. Take, for example, the rule that your specific spam message might try to get past if there is an image in the email (spammers tend to know how to use beacons):
HTML_IMAGE_ONLY_32: HTML: images with 2800-3200 bytes of words
(I have seen very similar emails to the one you mentioned with a beacon - a 1x1 invisible png hosted on a remote server, used to know if you've opened the email)
The specific one is not necessarily a case of Bayesian poisoning as only terrible spamfilters register a 5-line-long word as a word. It is simply attempting to buffer content - as some spamfilter rules work on the "very short (unsolicited) email = bad" rule.
In addition, certain spamfilters are also configured to dilute the penalty scores (spampoints). This is not the case for most servers, as it is typically a very bad idea, due to the very reason highlighted above: by padding enough "good" words, you can turn a +10 spam score into a +0.5 spam score. This is trivial - copy-paste from wikipedia.